PimSpace

App development

阅读中文版

Custom apps and SaaS, built end to end

PimSpace designs and builds custom software for Australian businesses and founders that need a specialised app: mobile apps for iOS and Android, SaaS tools and web apps, and the AI agents and integrations inside them. In English and Mandarin, from Sydney.

What we build

  • Mobile apps

    Apps for iOS and Android, from initial concept to production deployment.

  • SaaS tools and web apps

    Web apps and multi-tenant SaaS, built to run in production.

  • AI agents and integrations

    RAG, tool-calling agents and workflow automation inside the product, connected to the systems you already use.

How we build: the AI software factory

Giving developers an AI assistant does not make a business faster. The operating model does. Ours is a software factory: a team of AI agents, each with one job, moves every feature round the same loop, and a person makes the calls that matter. That is how we ship in weeks what used to take months, without cutting corners on quality or security.

AI agentteamSpecBuildReviewVerifyShipLearn
  1. 01

    Spec

    Every job starts as a written spec, with the checks that will prove it works. No spec, no build.

  2. 02

    Build

    A builder agent writes the code in its own workspace, in small steps it commits one at a time.

  3. 03

    Review

    A different agent, on a different model, reviews every change. No agent marks its own work.

  4. 04

    Verify

    A verifier agent runs the real app, checks each point in the spec and records the evidence.

  5. 05

    Ship

    Green checks merge on their own. A person approves anything that touches production or money.

  6. 06

    Learn

    Every miss becomes a rule the next run starts with. Shipping is one stop on the loop, not the end of it.

What the factory runs on

  • Linear

    The order book. Every job is a ticket with a spec.

  • Orca

    The factory floor. Agents work side by side, each in its own workspace.

  • Claude, Codex, Gemini, Droid

    The agent team. They plan, build, review and verify, and a different model checks each piece of work.

  • GitHub

    The quality gate. Nothing merges until the checks pass.

  • A person

    Decisions reach George's phone. One tap, and the line moves again.

In our pilot, a new feature went from ticket to done in two and a half hours, with one human click.

Early pilot: two apps, a handful of milestones.

Video thumbnail: The AI software factory

Watch

The AI software factory in 3 minutes

How agents ship features while the team sleeps: the stack, a real run from ticket to done, and what it means for IT leaders.

Watch on YouTube

Some of our work

Five systems George Qiao designed and built end to end. They are PimSpace's own products, not client work, and they show what the factory delivers. Three are open source.

Launch film. The restaurant is a demo; the voice and lifestyle shots are AI-generated.

Restaurant SaaS

OrderTogether

Guests scan the table's QR code and order together from their own phones, in their own language. Every phone updates live, and one tap sends the round to the kitchen screen.

How it works

Each guest can change only their own unsent items. Staff void an item with a reason, answer help calls from the floor view, and close the table on every phone at once. Each restaurant sees only its own data.

  • 11 languages
  • Real time on Convex
GreatJob public home: "A little help. A great job." with Create account and Sign in buttons beside a sunlit courtyard.

Public home of the beta build. For now, greatjob.au takes launch sign-ups.

Two-sided marketplace

GreatJob

Customers describe a job and get up to three quotes from local providers, free. They pay through a protected payment, and the provider is paid once the work is done.

How it works

Providers pay only when they earn, and rebooking the same provider costs less than the first job. Refunds and disputes are decided by a person, with an audit trail.

  • greatjob.au
  • Up to 3 quotes
  • Australia-wide

Recorded demo. Messages are sample FMCG complaints, not a live queue.

Real-time systems

Care Desk Live

A customer message lands in MongoDB. A change stream wakes a worker. Jev makes four judgments in about 300 ms, and a second change stream paints the verdict on every open screen. Nothing polls.

How it works

The card shows up as "Jev is reading", then gets a topic, a mood, an urgency, and a safety flag. A safety line becomes an escalation without a refresh.

  • About 300 ms
  • 10 topics
  • Two change streams
UniformOrder school list: Hawthorn Grammar, Ridgehaven School, and Riverside Academy.

1 of 4

Demo school. The names on the packing board are synthetic.

In production

UniformOrder

Parents order school uniforms on a phone. Volunteers pack paid orders from a tablet. Card payments settle to that school's own Stripe account.

How it works

One login covers two children at two schools. The board only shows orders that are already paid. The treasurer exports a GST-aware CSV instead of rebuilding a spreadsheet.

  • uniformorder.online
  • Stripe Connect per school
  • MIT
The room's mood stage with a join QR code, three words on screen, and an upbeat mood gauge.

The public stage. The words were already in the room.

One evening

The room's mood

People scan a QR code, type one word, and the stage shows each word as it lands. Claude Haiku scores the room. The count and the mood gauge move without a refresh.

How it works

Two screens, shipped the same night, with no private packages. The stage is for the room. The phone screen is one name and one word.

  • One night
  • Next.js 16
  • Claude Haiku

More will come.

Follow on GitHub

Bring us the hard one.

We don't take every app. We take the ones that are hard: many moving parts, real money, live data, or AI that has to be right. If yours has stalled, or another team called it too complex, that is where we start.

Tell us the hard part

Need marketing run instead? See how we run it